Compound Screening with Deep Learning for Neglected Diseases: Leishmaniasis

Jonathan A. J. Smith, Hao Xu, Xinran Li, Laurence Yang, Jahir GutierrezbioRxiv 2021 · DOI

TL;DR

This paper presents a deep learning pipeline for drug repurposing to combat leishmaniasis, a neglected tropical disease. By combining deep learning models with traditional molecular docking, they identified promising drug candidates, including Lacosamide, for further investigation.

What Problem Does This Paper Address?

Leishmaniasis, a neglected tropical disease affecting millions annually, suffers from a severe lack of effective and accessible treatments. Current therapies are often difficult to obtain, have adverse side effects, and are not optimally delivered. The development of new drugs is prohibitively expensive and time-consuming, with only a tiny fraction of research funding dedicated to these diseases. Drug repurposing, the process of identifying existing drugs for new therapeutic uses, offers a more cost-effective and faster alternative. However, efficiently screening vast libraries of existing compounds for efficacy against neglected diseases like leishmaniasis remains a significant challenge.

Methods

The researchers developed a multi-stage pipeline for drug repurposing. First, they employed deep learning models, specifically a multi-objective neural network (MONN) and the DeepPurpose framework, to predict binding affinities between potential drug compounds and Leishmania protein targets. These models were trained on existing protein-compound interaction datasets. Following the deep learning predictions, promising drug-protein pairs were further evaluated using traditional molecular docking simulations (AutoDock Vina) to refine binding potential and ranking. Finally, they performed clustering analysis on molecular descriptors of candidate drugs to identify subgroups with similar properties, aiding in the selection of top candidates.

Datasets

  • Protein Targets: Provided by the Indaba Grand Challenge, including sets for Leishmania major and general Leishmania targets.
  • Protein-Compound Interaction Data:
    • PDBBind v2018: Used for training the MONN model.
  • BindingDB: Used for training the DeepPurpose framework and for evaluating binding affinity metrics.
  • DrugBank Database: Used to identify approved and investigational drugs for further analysis.

Key Insights and Ideas

  • Deep Learning for Neglected Diseases: The study demonstrates the potential of deep learning to significantly accelerate and improve the efficiency of drug repurposing for neglected diseases, a field historically underserved by research funding.
  • Synergy of Deep Learning and Docking: Combining deep learning's ability to screen large datasets with the detailed physical insights of molecular docking provides a robust approach for identifying high-confidence drug candidates.
  • Lacosamide as a Promising Candidate: The identification of Lacosamide, an existing anti-seizure medication, as a strong candidate for treating leishmaniasis is a significant finding, highlighting the power of repurposing.
  • Molecular Descriptors for Understanding: Clustering molecules based on their descriptors helps to understand shared properties among effective drugs, potentially guiding future drug design or selection.

Key Results

  • The pipeline achieved a high ranking (2nd place) in the Indaba Grand Challenge for curing leishmaniasis.
  • Lacosamide was identified as a top drug candidate with a strong predicted binding score (-31.751 Indaba Score) against the Leishmania enzyme sterol 14-alpha demethylase.
  • The DeepPurpose model, specifically using the Daylight-AAC encoder, showed the highest correlation with the Indaba Scores (PyRosetta docking), indicating its effectiveness in this context.
  • The study identified several other promising drug candidates, though many are still under investigation or experimental.

Limitations and Future Work

  • Limited Data for Specific Organisms: The MONN model, which utilizes full 3D protein structures, was limited by the availability of such data for Leishmania species, potentially impacting its correlation with scoring metrics compared to DeepPurpose.
  • In Vivo Validation Needed: The identified drug candidates require further experimental validation, particularly through in vivo analysis, to confirm their efficacy, safety, and delivery constraints.
  • Expanding to Other Diseases: The authors suggest that the developed pipeline and approach can be expanded to address other neglected diseases, offering a broader impact on global health.
  • Further Exploration of Descriptors: Continued investigation into molecular descriptors that correlate with binding affinity could lead to more refined drug selection and design strategies.

Summary updated

Abstract

A bstract Deep learning provides a tool for improving screening of candidates for drug re-purposing to treat neglected diseases. We show how a new pipeline can be developed to address the needs of repurposing for Leishmaniasis. In combination with traditional molecular docking techniques, this allows top candidates to be selected and analyzed, including for molecular descriptor similarity.
BibTeX
@misc{smith2021compound,
  author = {Jonathan A. J. Smith and Hao Xu and Xinran Li and Laurence Yang and Jahir Gutierrez},
  title = {{Compound Screening with Deep Learning for Neglected Diseases: Leishmaniasis}},
  year = {2021},
  doi = {10.1101/2021.10.02.462874},
  howpublished = {bioRxiv},
}
Figure